big: implemented core shift routines in arith.go and

provide assembly versions (for x86-64 for now)

(Not yet used - waiting for previous CL to clear)

R=rsc
CC=golang-dev
https://golang.org/cl/1040041

src/pkg/big/arith.go

@@ -310,6 +310,28 @@ func subVW_g(z, x *Word, y Word, n int) (c Word) {
 }
 
 
+func shlVW(z, x *Word, s Word, n int) (c Word)
+func shlVW_g(z, x *Word, s Word, n int) (c Word) {
+	ŝ := _W - s
+	for i := 0; i < n; i++ {
+		w := *x.at(i)
+		c, *z.at(i) = w>>ŝ, w<<s|c
+	}
+	return
+}
+
+
+func shrVW(z, x *Word, s Word, n int) (c Word)
+func shrVW_g(z, x *Word, s Word, n int) (c Word) {
+	ŝ := _W - s
+	for i := n - 1; i >= 0; i-- {
+		w := *x.at(i)
+		c, *z.at(i) = w<<ŝ, w>>s|c
+	}
+	return
+}
+
+
 func mulAddVWW(z, x *Word, y, r Word, n int) (c Word)
 func mulAddVWW_g(z, x *Word, y, r Word, n int) (c Word) {
 	c = r

src/pkg/big/arith_386.s

@@ -99,6 +99,20 @@ E4:	CMPL BX, BP         // i < n
 	RET
 
 
+// func shlVW(z, x *Word, s Word, n int) (c Word)
+// TODO(gri) implement this routine
+TEXT ·shlVW(SB),7,$0
+	NOP			// work around bug in linker
+	JMP ·shlVW_g(SB)
+
+
+// func shrVW(z, x *Word, s Word, n int) (c Word)
+// TODO(gri) implement this routine
+TEXT ·shrVW(SB),7,$0
+	NOP			// work around bug in linker
+	JMP ·shrVW_g(SB)
+
+
 // func mulAddVWW(z, x *Word, y, r Word, n int) (c Word)
 TEXT ·mulAddVWW(SB),7,$0
 	MOVL z+0(FP), DI

src/pkg/big/arith_amd64.s

@@ -101,6 +101,56 @@ E4:	CMPQ BX, R11         // i < n
 	RET
 
 
+// func shlVW(z, x *Word, s Word, n int) (c Word)
+TEXT ·shlVW(SB),7,$0
+	MOVQ z+0(FP), R10
+	MOVQ x+8(FP), R8
+	MOVQ s+16(FP), CX
+	MOVL n+24(FP), R11
+	MOVQ $0, AX         // c = 0
+	MOVQ $0, BX         // i = 0
+	JMP E8
+
+L8:	MOVQ (R8)(BX*8), DX
+	MOVQ DX, R12
+	SHLQ CX, DX:AX
+	MOVQ DX, (R10)(BX*8)
+	MOVQ R12, AX
+	ADDL $1, BX          // i++
+
+E8:	CMPQ BX, R11         // i < n
+	JL L8
+
+	MOVQ $0, DX
+	SHLQ CX, DX:AX
+	MOVQ DX, c+32(FP)
+	RET
+
+
+// func shrVW(z, x *Word, s Word, n int) (c Word)
+TEXT ·shrVW(SB),7,$0
+	MOVQ z+0(FP), R10
+	MOVQ x+8(FP), R8
+	MOVQ s+16(FP), CX
+	MOVL n+24(FP), BX   // i = n
+	MOVQ $0, AX         // c = 0
+	JMP E9
+
+L9:	MOVQ (R8)(BX*8), DX
+	MOVQ DX, R12
+	SHRQ CX, DX:AX
+	MOVQ DX, (R10)(BX*8)
+	MOVQ R12, AX
+
+E9:	SUBL $1, BX         // i--
+	JGE L9
+
+	MOVQ $0, DX
+	SHRQ CX, DX:AX
+	MOVQ DX, c+32(FP)
+	RET
+
+
 // func mulAddVWW(z, x *Word, y, r Word, n int) (c Word)
 TEXT ·mulAddVWW(SB),7,$0
 	MOVQ z+0(FP), R10

src/pkg/big/arith_arm.s

@@ -18,6 +18,12 @@ TEXT ·addVW(SB),7,$0
 TEXT ·subVW(SB),7,$0
 	B ·subVW_g(SB)
 
+TEXT ·shlVW(SB),7,$0
+	B ·shlVW_g(SB)
+
+TEXT ·shrVW(SB),7,$0
+	B ·shrVW_g(SB)
+
 TEXT ·mulAddVWW(SB),7,$0
 	B ·mulAddVWW_g(SB)
 

src/pkg/big/arith_test.go

@@ -147,6 +147,36 @@ var prodVW = []argVW{
 	argVW{nat{_M << 7 & _M, _M, _M, _M}, nat{_M, _M, _M, _M}, 1 << 7, _M >> (_W - 7)},
 }
 
+var lshVW = []argVW{
+	argVW{},
+	argVW{nat{0}, nat{0}, 0, 0},
+	argVW{nat{0}, nat{0}, 1, 0},
+	argVW{nat{0}, nat{0}, 20, 0},
+
+	argVW{nat{_M}, nat{_M}, 0, 0},
+	argVW{nat{_M << 1 & _M}, nat{_M}, 1, 1},
+	argVW{nat{_M << 20 & _M}, nat{_M}, 20, _M >> (_W - 20)},
+
+	argVW{nat{_M, _M, _M}, nat{_M, _M, _M}, 0, 0},
+	argVW{nat{_M << 1 & _M, _M, _M}, nat{_M, _M, _M}, 1, 1},
+	argVW{nat{_M << 20 & _M, _M, _M}, nat{_M, _M, _M}, 20, _M >> (_W - 20)},
+}
+
+var rshVW = []argVW{
+	argVW{},
+	argVW{nat{0}, nat{0}, 0, 0},
+	argVW{nat{0}, nat{0}, 1, 0},
+	argVW{nat{0}, nat{0}, 20, 0},
+
+	argVW{nat{_M}, nat{_M}, 0, 0},
+	argVW{nat{_M >> 1}, nat{_M}, 1, _M << (_W - 1) & _M},
+	argVW{nat{_M >> 20}, nat{_M}, 20, _M << (_W - 20) & _M},
+
+	argVW{nat{_M, _M, _M}, nat{_M, _M, _M}, 0, 0},
+	argVW{nat{_M, _M, _M >> 1}, nat{_M, _M, _M}, 1, _M << (_W - 1) & _M},
+	argVW{nat{_M, _M, _M >> 20}, nat{_M, _M, _M}, 20, _M << (_W - 20) & _M},
+}
+
 
 func testFunVW(t *testing.T, msg string, f funVW, a argVW) {
 	n := len(a.z)
@@ -174,6 +204,18 @@ func TestFunVW(t *testing.T) {
 		testFunVW(t, "subVW_g", subVW_g, arg)
 		testFunVW(t, "subVW", subVW, arg)
 	}
+
+	for _, a := range lshVW {
+		arg := a
+		testFunVW(t, "shlVW_g", shlVW_g, arg)
+		testFunVW(t, "shlVW", shlVW, arg)
+	}
+
+	for _, a := range rshVW {
+		arg := a
+		testFunVW(t, "shrVW_g", shrVW_g, arg)
+		testFunVW(t, "shrVW", shrVW, arg)
+	}
 }